Black Forest Labs releases 'FLUX.3 Image Klein' (5.6B), an open-weight image generation and editing model, with Diffusers support

Verified (fact)

Black Forest Labs released 'FLUX.3 Image Klein', a FLUX.3-generation model, as open weights on Hugging Face, and Hugging Face announced Diffusers support. The 5.6B-parameter model generates images from text or edits existing images. It takes a prompt plus up to 10 reference images to produce a single image, handling text-to-image generation, single-reference editing and multi-reference composition in the same model. It uses Qwen3-VL-4B-Instruct as its text encoder.

Why it matters (interpretation)

The new generation of FLUX, a widely used open image model, has debuted first in a relatively small size. Because multi-reference editing is handled by a single model, it becomes easier to build image pipelines locally or on self-hosted servers. License terms could not be confirmed from the collected materials.

Sources

• Official Welcome FLUX.3 Image Klein in Diffusers Hugging Face
• Community Flux.3 Image Klein is now out Reddit